Recipes
Recipe: Migrating from Supabase
Este conteúdo não está disponível em sua língua ainda.
Overview
Section titled “Overview”This moves a Supabase project (tables, row-level security, SQL functions and triggers, users, a storage bucket, edge functions) to a Rebase project, locally first and then on Rebase Cloud. Everything is a CLI command. To keep running on the Supabase database instead of leaving it, see Supabase on the existing-database page.
The order matters: generate the tables from the source, then point the project at the target before anything writes.
1. Generate the tables
Section titled “1. Generate the tables”Introspect the Supabase database, or a local copy of it. SUPABASE_DB_URL
below is its direct connection string, as the postgres role (Supabase’s
dashboard: Connect → Direct connection); set it in your shell first:
SUPABASE_DB_URL="postgresql://postgres:<password>@db.<project-ref>.supabase.co:5432/postgres"rebase schema introspect --database-url "$SUPABASE_DB_URL" --from supabase--from supabase translates what a Rebase database does not have:
| In Supabase | Written as |
|---|---|
owner_id uuid default auth.uid() |
col.text().onCreateUser().userId(): the server writes the caller’s id |
owner_id uuid references auth.users(id) |
col.text().userId(), and the foreign key listed under “Not carried over” |
auth.uid() in a policy |
rebase.uid() |
TO authenticated / TO anon |
the role is dropped and the rule requires a signed-in / signed-out caller |
a policy only TO service_role |
dropped: server code runs in the server’s own context |
| plain indexes | the table’s indexes |
User-id columns become text because rebase.users.id and rebase.uid() are
text: a uuid column compared with rebase.uid() in a policy fails.
The run ends with a list of what it did not carry over (functions,
triggers, extensions, CHECK constraints, other indexes, composite foreign keys)
and writes the SQL for each to backend/sql/introspected-left-behind.sql.
2. Point the project at the new database
Section titled “2. Point the project at the new database”DATABASE_URL still names Supabase after step 1, and the introspect output
says so. Change it (or remove it, to use the managed development database)
before the first push:
rebase db push3. Apply the SQL objects
Section titled “3. Apply the SQL objects”Edit backend/sql/introspected-left-behind.sql the way the tables were
edited: auth.uid() → rebase.uid(), auth.users → rebase.users, no
GRANT … TO anon, authenticated. Functions that called pg_net are better as a
custom function. Then apply it:
psql "$DATABASE_URL" -f backend/sql/introspected-left-behind.sql# on Rebase Cloud, with no local psql:rebase cloud db query --file backend/sql/introspected-left-behind.sqlA function the browser called with supabase.rpc('fn') is reached through a
custom function that validates its input and calls it, as the caller under
row-level security:
// backend/functions/rpc.tsimport { defineFunction, getSql } from "@rebasepro/server/functions";
export default defineFunction((app) => { app.post("/get_public_proposal", async (c) => { const { token } = await c.req.json<{ token: string }>(); const sql = getSql(c); const [row] = await sql("select public.get_public_proposal($1::uuid) as result", { params: [token] }); return c.json(row?.result ?? null); });});4. Copy the data, keeping user ids
Section titled “4. Copy the data, keeping user ids”Keep the Supabase user ids, so every owner_id still points at its user:
insert the users first, with their ids, then the application tables. With
pg_dump against the source and the target. TARGET_URL is the new
database’s connection string: your DATABASE_URL, or on Rebase Cloud the
connectionString that rebase cloud db connect --reveal prints while it runs.
SUPABASE_DB_URL="postgresql://postgres:<password>@db.<project-ref>.supabase.co:5432/postgres"TARGET_URL="postgresql://…" # the new database: your DATABASE_URL, or the tunnel's URL on Rebase Cloudpsql "$SUPABASE_DB_URL" -c "\copy (select id::text, email, created_at from auth.users) to users.csv csv"psql "$TARGET_URL" -c "\copy rebase.users (id, email, created_at) from users.csv csv"pg_dump "$SUPABASE_DB_URL" --data-only --schema=public --no-owner > data.sqlpsql "$TARGET_URL" -f data.sqlPasswords do not move. Users sign in again with a magic link or a password reset.
5. Storage
Section titled “5. Storage”A public Supabase bucket maps to a storage source
with paths under public/, which are served without a token. Copy the objects
with any S3 client, keeping their keys.
6. Edge functions
Section titled “6. Edge functions”A Deno edge function becomes a custom function
in backend/functions/. The usual substitutions:
| Supabase | Rebase |
|---|---|
Deno.serve(handler) |
defineFunction((app) => app.post("/", handler)) |
Deno.env.get("X") |
requireEnv(c, "X") inside the handler |
createClient(url, serviceRoleKey) |
rebase.dataAsAdmin or rebase.sql on the server |
| a client with the user’s JWT | getSql(c): SQL as the caller, under row-level security |
EdgeRuntime.waitUntil(p) |
waitUntil(c, p) (background work) |
Supabase lets an edge function run for 150 s; Rebase’s default is 30 s. Raise
REBASE_FUNCTIONS_TIMEOUT_MS for one that calls an LLM.
7. Sign-in email
Section titled “7. Sign-in email”Magic-link and email-code sign-in need an email service. Set SMTP_HOST,
SMTP_PORT, SMTP_USER, SMTP_PASS and SMTP_FROM (rebase cloud env set
on Cloud). Without them the route answers 503, and rebase cloud status
reports blockedOn: email_not_configured.
8. Deploy
Section titled “8. Deploy”rebase cloud projects create --name Triptelier --linkrebase cloud deployrebase cloud db query --file backend/sql/introspected-left-behind.sql